An international comparison of individual tree detection and extraction using airborne laser scanning

Remo e Sens. 2012, 4, 950-974; doi:10.3390/ s4040950 Remo e Sensing ISSN 2072-4292 www.mdpi.com/jou nal/ emo esensing A icle An In e na ional Compa ison o Indi idual T ee De ec ion and Ex ac ion Using Ai bo ne Lase Scanning Ha i Kaa inen 1,*, Juha Hyyppä 1, Xiaowei Yu 1, Mikko Vas a an a 2, Hannu Hyyppä 3, An e o Kukko 1, Ma kus Holopainen 2, Ch is ian Heipke 4, Manuela Hi schmugl 5, Felix Mo sdo 6, E ik Næsse 7, Juho Pi känen 8, So in Popescu 9, S ein Solbe g 10, Be nd Michael Wol 11 and Jee-Cheng Wu 12 1 Depa men o Remo e Sensing and Pho og amme y, Finnish Geode ic Ins i u e, P.O.Box 15, FI-02431 Masala, Finland; E-Mails: juha.hyyppa@ gi. i (J.H.); [email p o ec ed] (X.Y.); [email p o ec ed] (A.K.) 2 Depa men o Fo es Sciences, Uni e si y o Helsinki, P.O. Box 27 (La oka anonkaa i 7), FI-00014 Helsinki, Finland; E-Mails: [email p o ec ed] (M.V.); [email p o ec ed] (M.H.) 3 School o Science and Technology, Aal o Uni e si y, FI-00076 Aal o, Finland; E-Mail: [email p o ec ed] 4 Ins i u e o Pho og amme y and GeoIn o ma ion, Leibniz Uni e si ä Hanno e , Nienbu ge S . 1, D-30167 Hanno e , Ge many; E-Mail: [email p o ec ed]anno e .de 5 Ins i u e o In o ma ion and Communica ion Technologies, Joanneum Resea ch Fo schungsgesellscha mbH, S ey e gasse 17, A-8010 G az, Aus ia; E-Mail: manuela.hi [email protected] 6 Depa men o Geog aphy, Uni e si y o Zü ich, Win e hu e s asse 190, CH-8057 Zü ich, Swi ze land; E-Mail: [email p o ec ed] 7 Depa men o Ecology and Na u al Resou ce Managemen , No wegian Uni e si y o Li e Sciences, P.O. Box 5003, NO-1432 Ås, No way; E-mail: e ik.naesse @umb.no 8 The Finnish Fo es Resea ch Ins i u e, P.O. Box 68 (Yliopis oka u 6), FI-80101 Joensuu, Finland; E-Mail: juho.pi kanen@me la. i 9 Depa men o Ecosys em Science and Managemen , Texas A&M Uni e si y, 2120 TAMU, College S a ion, TX 77843, USA; E-mail: s-popescu@ amu.edu 10 Depa men o Fo es Resou ces, No wegian o es and landscape ins i u e, P.O. Box 115, NO-1431 Ås, No way; E-Mail: [email p o ec ed] 11 Sol ing3D GmbH, Os e iede 8-10, D-30027 Ga bsen, Ge many; E-Mail: [email p o ec ed] 12 Depa men o Ci il Enginee ing, Na ional I-Lan Uni e si y, No. 1, Sec. 1, Sheng-Lung Road, I-Lan Ci y 260, Taiwan; E-Mail: [email p o ec ed] * Au ho o whom co espondence should be add essed; E-Mail: ha[email p o ec ed]; Tel.: +358-9-2955-5202. OPEN ACCESS Remo e Sens. 2012, 4 951 Recei ed: 10 Feb ua y 2012; in e ised o m: 15 Ma ch 2012 / Accep ed: 15 Ma ch 2012 / Published: 30 Ma ch 2012 Abs ac : The objec i e o he “T ee Ex ac ion” p ojec o ganized by Eu oSDR (Eu opean Spa ial da a Resea ch) and ISPRS (In e na ional Socie y o Pho og amme y and Remo e Sensing) was o e alua e he quali y, accu acy, and easibili y o au oma ic ee ex ac ion me hods, mainly based on lase scanne da a. In he inal epo o he p ojec , Kaa inen and Hyyppä (2008) epo ed a high a ia ion in he quali y o he published me hods unde bo eal o es condi ions and wi h a ying lase poin densi ies. This pape summa izes he indings beyond he inal epo a e analyzing he esul s ob ained in di e en ee heigh classes. Omission/Commission s a is ics as well as neighbo hood ela ions a e aken in o accoun . Addi ionally, ou au oma ic ee de ec ion and ex ac ion echniques we e added o he es . Se e al me hods in his expe imen we e supe io o manual p ocessing in he dominan , co-dominan and supp essed ee s o eys. In gene al, as expec ed, he alle he ee, he be e he loca ion accu acy. The accu acy o ee heigh , a e emo ing g oss e o s, was be e han 0.5 m in all ee heigh classes wi h he bes me hods in es iga ed in his expe imen . Fo o es in en o y, minimum cu a u e-based ee de ec ion accompanied by poin cloud-based clus e de ec ion o supp essed ees is a solu ion ha dese es a en ion in he u u e. Keywo ds: ee de ec ion; ee ex ac ion; ai bo ne lase scanning; Eu oSDR; ISPRS; indi idual ee in en o y; 3D; c own delinea ion 1. In oduc ion The de elopmen o lase / ada anging measu emen s wi hou scanning and p ope a i ude con ol o o es in en o y in 1970s–1990s [1–11] p omo ed he applica ion o lase measu emen in o es y and led o he apid adap a ion o ai bo ne lase scanning (ALS) in o es in en o y. A i s , ALS was applied in de e mining o es e ain ele a ions [12,13]. This was immedia ely ollowed by s andwise mean heigh and olume es ima ion [14–16], based on he da a collec ed ia anging measu emen s, and e y soon ALS was applied o in en o ying, ocusing on indi idual ees [17–20] wi h he ad en o apid image p ocessing, ee species classi ica ion [21,22] and he measu emen o ee g ow h and de ec ion o ha es ed ees [23] based on bi- empo al da a se s. O e 10 yea s, he ex ac ion o o es a iables has been di ided in o wo ca ego ies: a ea-based in en o ies and in en o ies based on indi idual ees o g oups o ees. Concu en ly wi h hese de elopmen s, lase scanning has inc easingly p o ided he co e da a se o mapping au ho i ies. The poin densi y o lase scanning has inc eased cons an ly. In addi ion o being used in o es in en o y, ALS da a om o es ed a eas is used o pu poses such as ligh obs acle mapping, powe line mapping, i ual ci y isualiza ion and mapping, and elecommunica ion planning. Remo e Sens. 2012, 4 952 In he No dic coun ies he e ie al o s and cha ac e is ics (e.g., mean ee heigh , dominan heigh , mean diame e , s em numbe , basal a ea, and imbe olume), which a e needed in o es managemen planning, is cu en ly being eplaced by ALS-based in en o y me hodologies. As ega ds ope a ional o es in en o ies, he wo-s age p ocedu e using ALS da a and ield plo s, i.e., a ea-based app oach (ABA, [24]), has become common and a e e ence o o he in en o y me hodologies. The o emos ad an ages o he s a e-o - he-a ABA, when compa ed o adi ional s andwise ield in en o y (SWFI), a e g ea e p ecision in he p edic ion o o es a iables [25], sampling-based es ima ion o o es a iables wi h he possibili y o calcula e accu acy s a is ics, and (a leas in p inciple) ALS-based in en o y is no dependen on s and bounda ies. Mo eo e , cu en ALS da a acquisi ion and p ocessing cos s a e less han hose o adi ional SWFI me hods. The ALS-based o es in en o y me hodology based on indi idual ee de ec ion (ITD) has been widely s udied ecen ly, bu is no widely used in p ac ice, due o assumed p oblems ela ed o ee de ec ion unde a ious o es condi ions [26–28]. O he p oblems ela ed o he p ac ical use o ITD include he need o highe ALS poin densi y, which adds o he cos s and he amoun o da a ha would need o be s o ed, as well as inadequa e ee species iden i ica ion accu acy. The assumed main ad an age o ITD would be ha i p o ides ue s em dis ibu ion se ies, enabling be e p edic ions o imbe asso men s. S em dis ibu ions a e p edic ed in he ABA, causing inaccu acy in imbe asso men es ima es and o es alue [29]. Ano he ad an age o ITD is he educed amoun o expensi e ieldwo k compa ed o ha needed when applying he ABA app oach. The esul s ob ained o indi idual ee ex ac ion ha e a ied signi ican ly om s udy o s udy. Pe cen age o co ec ly delinea ed ees has anged om 40% o 93%, [17,21,29–36]. I was no known how much o his a ia ion was caused by he me hods and how much by o es condi ions, un il he in e na ional benchma king s udy “T ee Ex ac ion” (2005–2008) was ca ied ou . In o de o es he ee ex ac ion me hods using he same da a se s, he Eu opean Spa ial Da a Resea ch O ganiza ion (Eu oSDR) and he In e na ional Socie y o Pho og amme y and Remo e Sensing (ISPRS) ini ia ed he “T ee Ex ac ion” p ojec o e alua e he quali y, accu acy, and easibili y o au oma ed ee ex ac ion me hods based on ai bo ne lase scanne da a and digi al ae ial images. The p ojec was hos ed by he Finnish Geode ic Ins i u e (FGI). Twel e pa ne s om USA, Canada, No way, Sweden, Finland, Ge many, Aus ia, Swi ze land, I aly, Poland and Taiwan pa icipa ed in he es included in he “T ee Ex ac ion” p ojec . The pa ne s we e eques ed o ex ac ees using he gi en ALS and image da ase s. Ano he objec i e o he s udy was o ind ou how he poin densi y impac s on indi idual ee ex ac ion. The esul s we e published in he p ojec ’s inal epo [27]. The epo se s ou he accu acy o ee ex ac ion pe pa ne , bu i does no p esen a mo e de ailed analysis o he esul s. The inal epo showed ha he ex ac ion me hod is he main ac o a ec ing achie ed accu acy. When he lase poin densi y inc eased om 2 poin s o 8 poin s pe m2, he imp o emen in c own delinea ion accu acy was ma ginal. T ee de ec ion accu acy esul s om he e ogeneous o es s a e p esen ed in Pi känen e al. [37] whe e he de ec ion accu acy was only 40% (70% o dominan ees). Yu e al. [38] p esen ed an accu acy o 69% o ee de ec ion unde a ious o es condi ions (di e en o es densi ies, ages, si e ypes and ee species). Heinzel e al. [39] in oduced an app oach ha classi ies c own size in ad ance and uses his in o ma ion as p io knowledge o single- ee ex ac ion. C own size is classi ied om ae ial colo in a ed image ex u e wi h an imp o ed g ey-scale g anulome y ollowed by a c own size Remo e Sens. 2012, 4 953 adap ed wa e shed segmen a ion o single ees. The accu acy a ies be ween 64% and 88%. Vauhkonen e al. [40] es ed se e al algo i hms unde di e en ypes o o es s; Eucalyp us plan a ion in B azil, coni e ous and deciduous plo s in Ge many and mainly coni e ous plo s in No way and in Sweden. The ee de ec ion a e a ied be ween 54% and 86%. These esul s a e on a comple ely di e en scale om hose ob ained by Peuhku inen e al. [41], whe e ITD was ca ied ou in wo ma ked s ands (densi y ~465 s ems pe ha). The numbe o ha es able ees was unde es ima ed by only <3%, bu his esul may include some commission e o s, i.e., a single ee is segmen ed in o se e al segmen s, hus inc easing he numbe o de ec ed ees. Falkowski e al. [42] showed ha ac oss a ull ange o canopy condi ions in a mixed-species, s uc u ally di e se coni e o es in no he n Idaho, Uni ed S a es, he e ec o he canopy co e densi y is signi ican o ee de ec ion accu acy. Vas a an a e al. [43] combined au oma ed ITD and isual in e p e a ion o acqui e e e ence da a o ABA. They assumed ha addi ional isual in e p e a ion would signi ican ly enhance he accu acy o he de i ed plo -le el o es a iables and p o ide supe io esul s when used o ain he ABA, in con as o me e au oma ed ITD. Visual in e p e a ion imp o ed he accu acy o ITD alida ed a plo -le el as RMSE o s em olume dec eased om 32.1% o 28.6%. Howe e , he e was no imp o emen in ABA p edic ions. Vas a an a e al. [28] in es iga ed ITD e o sou ces, and hei e ec s on o es managemen planning calcula ions. The in es iga ed e o sou ces we e de ec ion o ees, e o s in ee heigh p edic ion and e o s in ee diame e p edic ion. The e ec s o hese e o s we e analyzed wi h Mon e Ca lo simula ions. The esul s showed ha he o emos e o sou ce in ITD is in ee de ec ion. This pape includes u he analyses o he esul s ob ained in he “T ee Ex ac ion” p ojec [27]. Fou addi ional me hods we e added o compa ison, namely (1) Local maxima (LM) inding, (2) Mul i-scale Laplacian o Gaussian, (3) Minimum cu a u e-based ee de ec ion and (4) LM inding wi h a ying window size. The LM inding me hod is a ela i ely easy and as implemen a ion, eady o ope a ional use, and ad anced me hods should p o ide be e esul s. Addi ionally, he esul s a e analyzed o a ious ee heigh s, and omission and commission e o s we e included. Finally, mo e conclusions a e d awn om he expe imen ’s esul s. Sec ion 2 desc ibes he da a se s used in he s udy. Sec ion 3 p o ides a b ie desc ip ion o he me hods used by he pa ne s and o he me hods used o he e alua ion. Sec ion 4 p esen s he esul s o he e alua ed pa ame e s and discussion. The accu acy o ee heigh and loca ion de e mina ion, c own delinea ion and he numbe o ex ac ed ees a e analyzed. Key conclusions a e gi en in Sec ion 5. 2. Ma e ial 2.1. S udy A ea The selec ed wo managed o es es si es, named Si e A and Si e B (Figu e 1), we e close o each o he in Sou he n Finland, abou 18 km wes o Helsinki. Tes si es we e e y di e se, pa ly la and pa ly s eep e ain (max ele a ion di e ence 45 m), wi h pa ches o mixed and mo e homogeneous ee species in a ious s ages o de elopmen . The main ee species on he wo si es we e Sco s pine Remo e Sens. 2012, 4 954 (Pinus syl es is), No way sp uce (Picea abies) and Sil e and Downy bi ches (Be ula sp.). Si e A was 2.6 ha and Si e B 5.8 ha in size. Figu e 1. Si e A (Le ) and Si e B (Righ ), ee heigh s shown as colo -coded canopy heigh model (CHM). 2.2. Da a P o ided o T ee Ex ac ion ALS da a (Table 1) wi h h ee poin densi ies (2, 4 and 8 poin s pe m2) was p o ided o he pa ne s. A digi al e ain model (DTM) wi h 0.5m g id spacing was calcula ed using he Te aScan-so wa e (algo i hm based on Axelsson [44]) and deli e ed as ASCII-g id. The aining da ase was measu ed wi h a o al s a ion and included species, loca ion, diame e a b eas heigh (DBH), and c own delinea ion (using 3–5 poin s pe c own) o 75 ees. 2.3. Re e ence Da a The e e ence da a we e collec ed by means o g ound su eys and e es ial lase scanning (TLS). RTK–GPS (Leica SR530) and o al s a ion (T imble 5602S DR200+) equipmen we e used o c ea e a ne wo k o g ound con ol poin s (GCP) o e he s udy si es. The 3D accu acy o he GCP’s was es ima ed o be 2–3 cm, which could be e i ied by epea ed RTK-GPS measu emen s (accu acy 1 cm + 1–2 ppm in plane and 1.5–2 cm + 2 ppm in heigh [45]) and o al s a ion obse a ions. The loca ions o sphe ical e e ence a ge s o a e es ial lase scanne (Fa o LS880HE) we e de e mined Remo e Sens. 2012, 4 955 wi h e e ence o o al s a ion measu emen s on he basis o he GCPs. Te es ial lase scanning was ca ied ou in 48 loca ions (Figu e 1) o ob ain lase poin co e age o all e e ence ees on i e es plo s, wo plo s on Si e A and h ee plo s on Si e B. Toge he he i e plo s co e ed an a ea o 0.48 ha (5.7% o o al es si e a ea). The e e ence da a included he loca ions and species o 352 ees and he heigh s o 254 ees (Table 2). Table 1. ALS da a collec ed om he s udy si es. Acquisi ion 29 June 2004 Ins umen Op ech ALTM 2033 Fligh al i ude 600 m Pulse equency 33,000 Hz Field o View ±9 deg ees Measu emen densi y 2 poin s pe m2 pe echo pe s ip Swa h wid h 185 m Mode Fi s and las pulse Table 2. Re e ence da a ee species dis ibu ion and heigh s. T ee Heigh (m) Species % Mean Max. S d. Sco s pine 20 11.2 22.9 7.3 No way sp uce 46 14.2 25.5 6.7 Bi ch 15 20.3 27.2 5.3 O he deciduous 19 10.2 25.6 7.1 Figu e 2. Example o modeled e e ence ees, op iew on he le and side iew on he igh . Whi e: 3D ee ec o s; C op a ea o side iew; Heigh eading o one ee. Remo e Sens. 2012, 4 956 The poin clouds o he indi idual e es ial lase scannings we e geo e e enced using sphe ical e e ence a ge s. Then each poin cloud was iangula ed in o a 3D-mesh. The 3D ec o s o indi idual scannings we e combined (Figu e 2), and ee pa ame e s we e measu ed manually using 3D CAD so wa e. The measu ed ee pa ame e s included ee unk loca ion, ee op loca ion, ee heigh , and c own delinea ion. T ee heigh s we e measu ed only o ees ha had he ee op clea ly isible in he da a. C own delinea ion was analyzed as c own co e ed a ea, so indi idual c owns we e no sepa a ed on dense a eas. The in ensi y images o he o iginal scannings we e used o de e mine ee species. 2.4. P oduced T ee Ex ac ion Resul s The pa ne s we e eques ed o ex ac ees using he gi en ma e ial. They we e ee o use any me hod. The pa ne s we e asked o p o ide he ee loca ions and heigh s, c own delinea ions, and heigh s o he c own base o he olumes o each ee ha hey could ex ac . Nine ou o he wel e pa ne s in he “T ee Ex ac ion” p ojec used solely lase scanning da a and we e included in his compa ison. Two pa ne s used bo h lase scanning da a and ai bo ne images and one used solely ai bo ne images. 3. Me hods 3.1. Me hods Used by he Pa ne s The me hods (Table 3) ha e been epo ed in de ail in Kaa inen and Hyyppä [27]. He e, a b ie desc ip ion o he key elemen s o he me hods is p o ided. Table 3. Me hod names and pa ne s. Me hod Pa ne Coun y De iniens De iniens AG Ge many FOI Swedish De ense Resea ch Agency Sweden Hanno e Leibniz Uni e si ä Hanno e Ge many Me la Finnish Fo es Resea ch Ins i u e Finland No way No wegian Fo es and Landscape Ins i u e and No wegian Uni e si y o Li e Sciences No way Ilan Na ional I-Lan Uni e si y Taiwan Texas Texas A&M Uni e si y USA Udine Uni e si y o Udine I aly Zü ich Uni e si y o Zü ich Swi ze land 3.1.1. Me hod De iniens Me hod De iniens was implemen ed in eCogni ion Expe so wa e using he poin densi y o 8 poin s pe m2. The me hod may be di ided in o ou main asks: (1) C ea ion o a o es mask, (2) Ini ial spli ing o he o es mask, (3) Spli ing he o es mask in o ee c owns and (4) Co ec ion o o e -spli c owns. A low-pass il e was applied o emo e small gaps and excessi e local minima and maxima. The c ea ion o he o es mask was pe o med by h esholding he canopy heigh model (CHM) images a he heigh o 2 m. CHM is compu ed as he di e ence be ween DTM and digi al Remo e Sens. 2012, 4 957 su ace model (DSM). The me hod used he highes poin /pixel in he objec as he seed and expands he seed o he c own bounda ies, which we e iden i ied by he posi i e di e ence be ween he cu en and he p oposed pixel. This was epea ed un il all a eas wi hin he cu en objec had been included in o new objec s. The di e ence equi ed o o m a bounda y was de ined wi h a h eshold. The h eshold was ini ially high and a bounda y was o med only whe e a la ge di e ence occu s, in his case a 1 m. The key o his me hod is he classi ica ion o objec s in o wo g oups, c owns and c own-clus e s. Those objec s iden i ied as c owns we e emo ed om u he spli ing i e a ions and only conside ed la e , while c own-clus e s we e p ocessed u he in he hope o sepa a ing he c owns con ained wi hin hem. 3.1.2. Me hod FOI Me hod FOI was ully au oma ed and had he ollowing s eps: (1) DSM c ea ion, (2) DTM c ea ion, (3) CHM c ea ion, (4) CHM il e ing using di e en Gaussian il e s esul ing in di e en images, (5) Segmen ing o he di e en images sepa a ely and selec ing he segmen chosen o a speci ic a ea by i ing a pa abolic su ace o he lase da a, (6) Es ima ion o he heigh s and c own diame e s o he iden i ied ees using he ele a ion da a and he a ea o he segmen [30]. The me hod wo ks on g idded da a, in his case 0.25 m × 0.25 m. 3.1.3. Me hod Hanno e Me hod Hanno e was based on a ee model wi h h ee geome ic pa ame e s (size, ci cula i y and con exi y o he ee c own). The p ocessing s a egy comp ised ou s eps. Fi s , a wide ange o DSM scale le els was c ea ed. The second s ep was segmen a ion, achie ed by applying a wa e shed ans o ma ion. In he hi d s ep, he bes hypo hesis o a c own om he o e lapping segmen s o all le els based on he ee model was selec ed. The selec ion o he bes hypo heses was achie ed wi h he help o uzzy unc ions o he ee model pa ame e s [46]. 3.1.4. Me hod Me la In Me hod Me la, a DSM pixel was conside ed o be a low, di e ing pixel, i a leas se en (su ace models om poin densi y o eigh poin s pe m2), o six (o he poin densi ies) o he eigh nea es neighbo s we e mo e han i e me e s highe han he pixel i sel . These pixels we e eplaced wi h he median o he mo e- han- i e-me e s-la ge neighbo pixel alues. The DTM was hen sub ac ed om he inal DSM o ge he CHM o he ee c own segmen a ion. Be o e segmen a ion, he CHM was smoo hed wi h heigh -based il e ing. Fi e Gaussian il e s we e used so ha he il e size inc eased along he heigh o he pixel being smoo hed. The minimum and maximum σ alues we e selec ed by isually ensu ing ha he numbe o LM was easonable a bo h ends o he ee heigh ange. A nega i e image o he heigh - il e ed image was hen c ea ed o wa e shed segmen a ion ha was used o sepa a e he ee c owns om each o he . Wa e shed egions associa ed wi h he local minima in he nega i e image we e iden i ied using an algo i hm ollowing he d ainage di ec ion. To de e mine he bounda ies be ween c owns and backg ound, pixels lowe han 2 m in he heigh - il e ed image we e masked ou om he c own segmen s. Finally, small segmen s (a mos h ee pixels in size) we e combined wi h one o he neighbo segmen s, his being a ee c own o he backg ound, based on he Remo e Sens. 2012, 4 958 smalles a e age g adien on he common segmen bounda y. The ee loca ions and heigh s we e hen ob ained om he loca ion and he alue o he pixel ha ing he highes alue wi hin each segmen . 3.1.5. Me hod No way In Me hod No way, he p ocess was comp ised o se e al s eps, i.e., e ain uppe mos echoes, in e pola e hem in o a DSM-g id, ind LM in he DSM, un a egion-g owing algo i hm wi h some es ic ions in o de o de i e objec s belonging o he class o objec s o en named s a objec s. The DSM was now di ided in o segmen s ha ep esen ed ee c owns, while pa s o he a ea we e no co e ed by ees and had no DSM alue. The DSM was adjus ed (li ed) using he esiduals be ween he DSM and he i s echoes. The 90 pe cen ile o he esiduals was calcula ed, and his equen ly u ned ou o be a ound 70 cm. This o se was added o all z alues in he DSM. The ee heigh s we e de i ed as he z alue o he LM a e his adjus men [47]. 3.1.6. Me hod Ilan In Me hod Ilan, he ee loca ions and ee heigh we e compu ed om he CHM. The CHM was compu ed by selec ing he highes lase poin wi hin each 1 m × 1 m g id cell. Fu he mo e, each unk o he aining da ase was loca ed wi hin a 3 m × 3 m window, and a heigh his og am wi h one-me e in e al was used o build up a lase classi ica ion ee model o species de e mina ion. The highes lase echo wi hin a speci ic a ea (i.e., 3 × 3 m), was assumed o be he po en ial unk loca ion o a ee. Two app oaches we e used o es ima e he po en ial ee loca ions. The i s app oach in ol ed unning a LM il e in he CHM wi h a window size o 3 × 3, and all po en ial ee loca ions we e selec ed. The second me hod p ocessed only heigh s less han 15 m in he CHM wi h 3 × 3 LM il e . The i s app oach was applied o Si e B, and he second app oach was applied o Si e A. The c own wid hs we e de i ed, based on he aining ee da a ( he ela ionship be ween ee species, ee heigh and c own wid h) and he CHM. The empi ical ela ionship be ween he heigh o he ees and hei sea ching c own size was de ined. 3.1.7. Me hod Texas Me hod Texas employed an au oma ed algo i hm o LM il e wi h a ci cula mo ing window o a ying sizes. The algo i hm was applied on CHMs wi h di e en pixel sizes depending on he poin densi ies: 50 cm and 100 cm o 2 poin s pe m2, 50 cm o 4 poin s pe m2, and 25 cm and 50 cm o 8 poin s pe m2. The LM il e is o en used o loca e ee posi ions based on he assump ion ha he highes ele a ion co esponds o he ee apex. When applying he LM il e , he window size has a majo impac on ee iden i ica ion. On he o he hand, i can be assumed ha he alle a ee is, he la ge is he wid h o i s c own. Thus, de e mining he il e size was based on he ela ionship be ween c own size and ee heigh . P io in o ma ion was u ilized o de i e such a ela ionship. To p edic he c own size, eg ession models we e i ed wi h ee heigh as he independen a iable. Fo he es da a, abou 90 ees on each si e we e isually iden i ied om he CHMs and he co esponding heigh s and c own wid hs we e manually eco ded by means o on-sc een measu emen . The c own diame e Remo e Sens. 2012, 4 965 applicable when he poin densi y is high and when he objec i e is o ocus only on he dominan o co-dominan ee laye s. In he FGI models he poin densi y imp o emen was no s udied, and FOI was he only me hod o imp o e he de ec ion o ees “in a g oup o simila ees” when he lase poin densi y inc eased. In he smalles ee classes, he lase poin densi y did no imp o e he de ec ion a e o hese ees. De iniens, Hanno e , and Texas beha ed so ha hey assumed ee c own size o be much la ge han he ac ual size and his esul ed in signi ican unde es ima ion o he numbe o ees, e en when dealing wi h he alles ee class. They did no include a p ope y o de ec ing he smalle ees unde nea h dominan ees. The e should be mo e ocus on inding smalle ees om unde nea h he dominan s o ey, e.g., he use o wa e o m echnology should be s udied. In p inciple, a highe pulse densi y should esul in a be e capabili y o inding ees, bu his also depends on he o es ype. I appea ed ha he es si es we e ai ly sui able o indi idual ee de ec ion wi h a poin densi y as low as 2 poin s pe m2. I is possible ha some o he po en ial imp o emen as a unc ion o poin densi y was los by c ea ing highe poin densi ies wi h epea ed lase s ips, each ha ing 2 poin s pe m2. Wi h epea ed lines, he lase poin dis ibu ion may no be op imal, e.g., when scanning lines a e close o each o he in each s ip. Figu e 6. The pe cen ages o ma ched and missed ees and he commission e o (ex a ees in he model). 0 10 20 30 40 50 60 70 80 90 100 [%] Ma ched and missed ees and commission e o Ma ch Missed bu in c own a ea Missed and ou side c own a ea Commission e o [%] shown o e he model name 4.2. Accu acy o De e mining T ee Loca ion The esul s on accu acy o ee loca ion (Figu e 7) clea ly showed ha he sensi i i y o ee loca ion is low as a unc ion o poin densi y and i mainly changes as a unc ion o he model p o ide . Ob iously, he calib a ion o he models has no been success ul and se e al models o e es ima ed he c own size (e.g., De iniens, Hanno e , and Zü ich). The de ec ed ee posi ions o he bes models show a s anda d de ia ion o less han 1m compa ed o he e e ence posi ions. Fo hese models, he Remo e Sens. 2012, 4 966 poin densi y has a negligible e ec on he esul ing ee posi ions. Fo ees highe han 15 m, a s anda d de ia ion o 0.5 m was ob ained o he ee posi ions. The au oma ed models we e as good as he manual p ocessing o he poin cloud in de e mining ee loca ions. Due o he me hod used o he e alua ion o ee loca ion, he capabili y e o s in ee inding a e included in he esul s, and he e o e loca ion accu acy as a unc ion o ee heigh is o in e es . FOI me hods appea o be sui able o loca ing ees, bu he esul s gi e an op imis ic iew o hem since hey ound mainly he “mo e easily con i mable cases”, i.e., alle and isola ed ees. Figu e 7. Accu acy o de ec ed ee loca ions. 0.0 0.5 1.0 1.5 2.0 De iniens FOI_2 FOI_4 FOI_8 Hanno e _2 Hanno e _4 Hanno e _8 Me la_2 Me la_4 Me la_8 No way_2 No way_4 No way_8 Ilan_2 Ilan_4 Ilan_8 Texas_2_100 Texas_2_50 Texas_4_50 Texas_8_50 Texas_8_25 Udine_2 Udine_4 Udine_8 Zu ich_2 Zu ich_4 Zu ich_8 Manual FGI_LOCM FGI_MLOG FGI_MCV FGI_VWS [m] Accu acy o de ec ed ee loca ions Mean STD RMSE Figu e 8. Accu acy o ee loca ions dependan on ee heigh classes. 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 [m] RMSE o ee loca ions dependan on ee heigh classes 2-5m 5-10m 10-15m 15-20m O e 20m T ee heigh classes: The esul s o he accu acy o ee loca ion by ee heigh a e shown in Figu e 8. In gene al, as was o be expec ed, he alle a ee is, he be e is he accu acy o loca ion. I should be emphasized ha he numbe o ees in he smalle ee classes was e y low (on an a e age only 7.5 ees in heigh class 2–5 m), and he accu acy alues should be ca e ully conside ed. Remo e Sens. 2012, 4 967 4.3. Accu acy o T ee Heigh The in es iga ed accu acy o he de ec ed ee heigh s shows ha di e en poin densi ies ha e a negligible e ec compa ed o he a iabili y be ween he me hods (Figu e 9). The bes models show a RMSE o 60 cm o 80 cm o he ob ained ee heigh s. The me hods o FOI, Me la, Texas, and FGI_VWS show he bes accu acies. The esul s wi h he bes au oma ed models we e signi ican ly be e han hose a ained when using he manual p ocess. In gene al, unde es ima ion o bo h ee heigh and s anda d de ia ion dec eased as he poin densi y inc eased. The o e es ima ion p oduced by he Model No way in ega d o ee heigh was due o he co ec ion applied o he ee heigh in he p ep ocessing phase. The me hods capable o inding mo e ees in he lowe classes ob iously su e ; he unce ain y ega ding he heigh s o he ex ac ed ee in he lowe le els is g ea e . Figu e 9. Accu acy o de ec ed ee heigh s. Figu e 10. Accu acy o de ec ed ee heigh s dependan on di e en heigh classes. 0.0 0.5 1.0 1.5 2.0 2.5 [m] RMSE o ee heigh s dependan on ee heigh classes 2-5m 5-10m 10-15m 15-20m O e 20m When he esul s o ee heigh accu acy as a unc ion o ee heigh we e compu ed, only ees wi h heigh di e ence be ween he e e ence and he model below 3 m we e used (based on he esul s Remo e Sens. 2012, 4 968 epo ed by Kaa inen and Hyyppä [27]). In he Hanno e model, he e o s we e so la ge wi h bigge ees ha ees belonging o heigh classes 15–20 m and o e 20 m we e omi ed om he analysis. Figu e 10 shows he esul s a e he 3m il e ing and dele ing o ou lie s. In gene al, ee heigh did no appea o impac he accu acy o heigh de e mina ion, al hough he me hods ha a e mos easible o smalle ee de ec ion su e he mos . Su p isingly, he me hods u ilizing LM inding a e among he bes ex ac ion echniques in ega d o ee heigh accu acy. 4.4. C own Delinea ion Accu acy The esul s o o al c own a ea appea o a y signi ican ly be ween he models (Figu e 11). The e o s leading o alse o al c own a ea a e as ollows: inadequa e ee inding capabili y (small ees missed), inadequa e il e ing o he aw poin cloud da a o DSM (leading o excessi ely la ge c owns, unde es ima ing he numbe o ees), and inadequa e calib a ion o he me hod using he gi en aining da a. Wi h me hods ha ep esen he ee c own using a simple o m such as a ci cle, i is possible ha when he poin densi y inc eases, he ee c own a ea also inc eases, as he e a e mo e hi s on he ou e b anches o he ees. As he e e ence delinea ion is a mo e de ailed han a ci cle, a ci cle o en o e es ima es he a ea al hough a ci cle i ed o a c own which was modeled using a highe poin densi y co esponds be e o he ou e b anches. This can be seen, o example, in he Udine model’s esul s. The models es ed mo e in p ac ical o es y in bo eal o es ha e al eady accumula ed mo e expe ience in his calib a ion, i.e., FGI, FOI, and No way me hods. Figu e 11. To al c own a ea accu acy. -50.0 -40.0 -30.0 -20.0 -10.0 0.0 10.0 20.0 30.0 40.0 50.0 De iniens FOI _ 2 FOI_4 FOI_8 Hanno e _ 2 Hanno e _4 Hanno e _ 8 Me la_2 Me la_4 Me la_8 No way_2 No wa y_ 4 No way_8 Ilan _ 2 Ilan_4 Ilan_8 Texas _ 2 _ 100 Texas_2_50 Texas _ 4 _ 50 Texas_8_50 Texas _ 8 _ 25 Udine_2 Udine_4 Udine _ 8 Zu ich_2 Zu ich _ 4 Zu ich_8 Manual FGI _ LOCM FGI_MLOG FGI _ MCV FGI_VWS [%] Compa ison o he es ima ed o al c own a ea o he e e ence 5. Conclusions The esul s con i med ha he ex ac ion me hod is he main ac o impac ing achie ed accu acy, as p oposed by Kaa inen and Hyyppä [27], and ha lase poin densi y has less impac on indi idual ee de ec ion. Remo e Sens. 2012, 4 969 Depending on he applica ion, he c i e ia o ee de ec ion and ex ac ion can be di e en . Fo example, he me hods ha a e mos sui able o dominan ee de ec ion should be used in ligh obs acle mapping, powe line moni o ing, and elecommunica ion planning. In his e alua ion, he bes me hods o dominan ee de ec ion we e FGI_MCV, FGI_LOCM, No way, and Me la, and FOI i he pulse densi y is close o 8 poin s pe m2. They we e also be e han manual p ocessing. Fo o es in en o y, he me hod should be such ha i is sui able o eco ding accu a ely he s and DBH dis ibu ion. Since DBHs a e calcula ed o each indi idual ee based on heigh , ee species and c own size, he o emos c i e ion is co ec heigh dis ibu ion. Allome ic models desc ibing he ela ionships be ween ee c own size, heigh , and DBH a e highly sensi i e o e o s in he inpu da a. Au oma ed measu emen esul s o ee c own size, in pa icula , end o be p one o e o s. Thus, he es ima ion o DBH on he basis o ee heigh and c own size esul s in a ai ly signi ican deg ee o unce ain y. Nea es -neighbo me hods applicable o single- ee in e p e a ion a e, he e o e, cu en ly unde de elopmen [23,50,51]. The ex ac ed da a, acqui ed om de ec ed ees, need o be calib a ed wi h he g ound u h, bu i is i al ha he me hod e eals as co ec ly as possible he numbe o dominan and supp essed ees wi h a small numbe o commission e o s. This is an aspec whe e signi ican wo k has ye o be unde aken. Ex ac ed indi idual ees can also be used in a simple way o imp o ing a ea-based es ima es wi h signi ican ly imp o ed accu acy and wi hou using any calib a ions [52], and hus indi idual ee ex ac ion echniques a e cu en ly also impo an om he p ac ical o es y poin o iew. Based on he esul s o he p esen expe imen , an app oach using he high de ec ion a e o FGI_MCV o all ee sizes (bu u he p ocessing he supp essed ee da a a he poin le el, as is done in he Zü ich app oach) is a solu ion dese ing o mo e a en ion. Full wa e o m echnology is also expec ed o imp o e indi idual ee de ec ion, especially in he case o supp essed ees, as wa e o m analysis can be used o p oduce dense poin clouds wi hin he c owns. Addi ionally, a simple me hod based on LM inding (p oposed, e.g., in Hyyppä e al. [20]) has u ned ou o be one o he bes echniques applied in his expe imen , and hanks o i s simple implemen a ion, i is especially easible o comme cial p oduc ion. In he ee clus e app oach (TCA), he i s phase is o segmen CHM, as is done in many ITD app oaches. In he second phase, accu a ely loca ed ield ees a e linked o he co esponding segmen s [53–56]. Con a y o he ITD model, i is no assumed ha a single segmen ep esen s a single ee. In he TCA, all he ield ees loca ed wi hin he segmen s’ a ea a e linked o co esponding segmen s. Thus, segmen s may include none, one, wo o e en mo e ees. This sol es he ee de ec ion p oblem in p ac ice and he bias o he es ima ed a ea-le el olume and basal-a ea is educed. Indi idual ee ex ac ion is pe haps one o he ew applica ions whe e au oma ion p o ides highe quali y han does manual p ocessing. Vas a an a e al. [43] ied o imp o e he au oma ically de ec ed ees wi h manual p ocessing when acqui ing e e ence o a ea-based in en o y, bu he accu acy o a ea-based p edic ions could no be imp o ed. Se e al o he me hods in he p esen expe imen we e supe io o manual p ocessing in dealing wi h dominan , co-dominan , and supp essed ee s o eys. This also means ha manually p ocessed ee maps based on ai bo ne lase su eys canno be used as e e ence o de eloping au oma ic algo i hms o ee de ec ion, al hough his has been done p e iously. In en o ies based on indi idual ees equi e e e ence da a o indi idual ees collec ed in Remo e Sens. 2012, 4 970 he ield by some o he means. Calib a ion is needed o educe he unde es ima ion o ee heigh and calib a ion o he basal a ea and s em olume (e.g., [57]). In gene al, and as was o be expec ed, he alle a ee is, he be e loca ion accu acy is, as could be expec ed. T ee heigh accu acy, a e high ou lie s had been dele ed, was be e han 0.5 m in all ee heigh classes when using he leading me hods in his expe imen , and his is a signi ican esul , e en hough he numbe o he lowes le el ees was small. Solu ions based on indi idual ees can be applied e en wi h poin densi ies o 2 poin s pe m2 o lowe (e.g., [57]), bu he op imum poin densi y is mos p obably dependen on ee size and s and densi y o he o es . When dealing wi h sapling s ands, a poin densi y o 10 poin s pe m2 o highe is expec ed o inc ease he accu acy o he ex ac ion esul s. This in e na ional benchma king expe imen also demons a ed ha he quali y o one me hod e sus o he me hods canno be e i ied wi hou es ing he me hods in he same o es condi ions, since he e ec o a iabili y o o es condi ions is belie ed o ha e a high impac on he achie ed accu acy. This is e iden when he esul s achie ed in his s udy a e compa ed o hose epo ed in exis ing li e a u e. Acknowledgmen s Au ho s wish o hank I is Lingen elde , Åsa Pe sson, Aiko Sukdolak, And ea Ba ilo i and F ancesco Sepic o coope a ion in he p ojec . F ançois Gougeon con ibu ed signi ican ly o he planning o he p ojec . Xinlian Liang, om FGI, also made signi ican con ibu ions. The suppo o FM-Ka a Oy (Blom Ka a Oy) in p o iding da a o he p ojec is g a e ully acknowledged. The Academy o Finland, in he o m o he p ojec s “Science and Technology Towa ds P ecision Fo es y” and “Towa ds Imp o ed Cha ac e iza ion o Map Objec s”, is acknowledged o inancial suppo . Re e ences 1. Solodukhin, V.I.; Zuko , A.J.; Mazugin, I.N. Lase ae ial p o iling o a o es . Lew. NIILKh. Lening ad. Lesnoe Khozyais o 1977, 10, 53-58 (in Russian). 2. Nelson, R.; K abill, W.; Maclean, G. De e mining o es canopy cha ac e is ics using ai bo ne lase da a. Remo e Sens. En i on. 1984, 15, 201-212. 3. Sch eie , H.; Lougheed, J.; Tucke , C.; Leckie, D. Au oma ed measu emen s o e ain e lec ion and heigh a ia ions using ai bo ne in a ed lase sys em. In . J. Remo e Sens. 1985, 6, 101-113. 4. Nelson, R.; K abill, W.; Tonelli, J. Es ima ing o es biomass and olume using ai bo ne lase da a. Remo e Sens. En i on. 1988, 24, 247-267. 5. Maclean, G.A.; K abill, W.B. G oss-me chan able imbe olume es ima ion using an ai bo ne lida sys em. Can. J. Remo e Sens. 1986, 12, 7-18. 6. Cu ie, D.; Shaw, V.; Be cha, F. In eg a ion o Lase Range inde and Mul ispec al Video Da a o Fo es Measu emen s. In P oceedings o IGARSS’89, Vancou e , BC, Canada, 10–14 July 1989; Volume 4; pp. 2382-2384. 7. Be na d, R.; Vidal-Madja , D.; Baudin, F.; Lau en , G. Nadi looking ai bo ne ada and possible applica ions o o es y. IEEE T ans. Geosci. Remo e Sens. 1987, 21, 297-309. Remo e Sens. 2012, 4 971 8. Hallikainen, M.; Hyyppä, J.; Some salo, E. Classi ica ion o Fo es Types by Mic owa e Remo e Sensing. In P oceedings o EARSeL 9 h Gene al Assembly and Symposium, Espoo, Finland, 27 June–1 July 1989; pp. 293-298. 9. Hyyppä, J.; Hallikainen, M.; Pulliainen, J. Accu acy o Fo es In en o y Based on Rada -De i ed S and P o ile. In P oceedings o IGARSS’93, Tokyo, Japan, 18–21 Augus 1993; pp. 391-393. 10. Hyyppä, J.; Hallikainen, M.; Hyyppä, H. A Scanning Ranging Rada o Fo es In en o y. In P oceedings o URSI/IEEE/IRC XXI Na ional Con en ion on Radio Science 1996, Espoo, Finland, 2–3 Oc obe 1996; Repo S 222; pp. 255-256. 11. Hyyppä, J.; Hyyppä, H.; Inkinen, M. Capabili ies o Mul i-Sou ce Remo e Sensing o Fo es In en o y. In P oceedings o 3 d In e na ional Ai bo ne Remo e Sensing Con e ence and Exhibi ion, Copenhagen, Denma k, 7–10 July 1997. 12. K aus, K.; P ei e , N. De e mina ion o e ain models in wooded a eas wi h ai bo ne lase scanne da a. ISPRS J. Pho og amm. 1998, 53, 193-203. 13. Vosselman, G. Slope Based Fil e ing o Lase Al ime y Da a. In P oceedings o XIX h ISPRS Cong ess Technical Commission III: Sys ems o Da a P ocessing, Analysis and Rep esen a ion, Ams e dam, The Ne he lands, 16–23 July 2000; In In e na ional A chi es o he Pho og amme y, Remo e Sensing and Spa ial In o ma ion Sciences; ISPRS: Vienna, Aus ia, 2000; Volume 33, No. B3/2, pp. 935-942. 14. Næsse , E. De e mina ion o mean ee heigh o o es s ands using ai bo ne lase scanne da a. ISPRS J. Pho og amm. 1997, 52, 49-56. 15. Næsse , E. Es ima ing imbe olume o o es s ands using ai bo ne lase scanne da a. Remo e Sens. En i on. 1997, 61, 246-253. 16. Hyyppä, H.; Hyyppä, J. Compa ing he accu acy o lase scanne wi h o he op ical emo e sensing da a sou ces o s and a ibu es e ie al. Pho og amm. J. Fin. 1999, 16, 5-15. 17. Hyyppä, J.; Inkinen, M. De ec ing and es ima ing a ibu es o single ees using lase scanne . Pho og amm. J. Fin. 1999, 16, 27-42. 18. B and be g, T. Au oma ic Indi idual T ee-Based Analysis o High Spa ial Resolu ion Remo ely Sensed Da a. Ph.D. Thesis, Ac a Uni e si a is Ag icul u ae Sueciae, Sil es ia 118, Swedish Uni e si y o Ag icul u al Sciences, Uppsala, Sweden, 1999. 19. Ziegle , M.; Kon ad, H.; Ho ich e , J.; Wimme , A.; Ruppe , G.; Scha d , M.; Hyyppä, J. Assessmen o o es a ibu es and single- ee segmen a ion by means o lase scanning. P oc. SPIE 2000, 4035, 73-84. 20. Hyyppä, J.; Scha d , M.; Hagg én, H.; Koch, B.; Loh , U.; Sche e , H.U.; Paananen, R.; Luukkonen, H.; Ziegle , M.; Hyyppä, H.; e al. HIGH-SCAN: The i s Eu opean-wide a emp o de i e single- ee in o ma ion om lase scanne da a. Pho og amm. J. Fin. 2001, 17, 58-68. 21. B and be g, T.; Wa ne , T.; Landenbe ge , R.; McG aw, J. De ec ion and analysis o indi idual lea -o ee c owns in small oo p in , high sampling densi y lida da a om he eas e n deciduous o es in No h Ame ica. Remo e Sens. En i on. 2003, 85, 290-303. 22. Holmg en, J.; Pe sson, Å. Iden i ying species o indi idual ees using ai bo ne lase scanning. Remo e Sens. En i on. 2004, 90, 415-423. Remo e Sens. 2012, 4 972 23. Yu, X.; Hyyppä, J.; Kaa inen, H.; Mal amo, M. Au oma ic de ec ion o ha es ed ees and de e mina ion o o es g ow h using ai bo ne lase scanning. Remo e Sens. En i on. 2004, 90, 451-462. 24. Næsse , E. P edic ing o es s and cha ac e is ics wi h ai bo ne scanning lase using a p ac ical wo-s age p ocedu e and ield da a. Remo e Sens. En i on. 2002, 80, 88-99. 25. Holopainen, M.; Vas a an a, M.; Rasinmäki, J.; Kallio i a, J.; Mäkinen, A.; Haapanen, R.; Melkas, T.; Yu, X.; Hyyppä, J. Unce ain y in imbe asso men es ima es p edic ed om o es in en o y da a. Eu . J. Fo . Res. 2010, 129, 1131-1142. 26. Falkowski, M.; Smi h, A.; Gessle , P.; Hudak, A.; Vie ling, L.; E ans, J. The in luence o he coni e o es canopy co e on he accu acy o wo indi idual ee de ec ion algo i hms using lida da a. Can. J. Remo e Sens. 2008, 34, 1-13. 27. Kaa inen, H.; Hyyppä, J. Eu oSDR/ISPRS P ojec , Commission II “T ee Ex ac ion”; Final Repo ; O icial Publica ion no 53; Eu oSDR (Eu opean Spa ial Da a Resea ch): Dublin, I eland, 2008. 28. Vas a an a, M.; Holopainen, M.; Yu, X.; Hyyppä, J.; Mäkinen, A.; Rasinmäki, J.; Melkas, T.; Kaa inen, H.; Hyyppä, H. E ec s o ALS indi idual ee de ec ion e o sou ces on o es managemen planning calcula ions. Remo e Sens. 2011, 3, 1614-1626. 29. Holopainen, M.; Mäkinen, A.; Rasinmäki, J.; Hyyppä, J.; Hyyppä, H.; Kaa inen, H.; Vii ala, R.; Vas a an a, M.; Kangas, A. E ec o ee le el ai bo ne lase scanning accu acy on he iming and expec ed alue o ha es decisions. Eu . J. Fo . Res. 2010, 29, 899-910. 30. Pe sson, Å.; Holmg en, J.; Söde man, U. De ec ing and measu ing indi idual ees using an ai bo ne lase scanne . Pho og amm. Eng. Remo e Sensing 2002, 68, 925-932. 31. Leckie, D.; Gougeon, F.; Hill, D.; Quinn, R.; A ms ong, L.; Sh eenan, R. Combined high-densi y lida and mul ispec al image y o indi idual ee c own analysis. Can. J. Remo e Sens. 2003, 29, 633-649. 32. S aub, B. A Top-Down Ope a o o he Au oma ic Ex ac ion o T ees: Concep and Pe o mance E alua ion. In P oceedings o he ISPRS Wo king G oup III/3 Wo kshop ‘3-D Recons uc ion om Ai bo ne Lase scanne and InSAR Da a’, D esden, Ge many, 8–10 Oc obe 2003; pp. 34-39. 33. Popescu, S.; Wynne, R.; Nelson, R. Measu ing indi idual ee c own diame e wi h lida and assessing i s in luence on es ima ing o es olume and biomass. Can. J. Remo e Sens. 2003, 29, 564-577. 34. Ande sen, H-E.; Reu ebuch, S.; Sch eude , G. Bayesian Objec Recogni ion o he Analysis o Complex Fo es Scenes in Ai bo ne Lase Scanne Da a. In P oceedings o ISPRS Commission III, Symposium 2002 Pho og amme ic Compu e Vision, G az, Aus ia, 9–13 Sep embe 2002; In In e na ional A chi es o he Pho og amme y, Remo e Sensing and Spa ial In o ma ion Sciences; ISPRS: Vienna, Aus ia, 2002; Volume 34, pa 3A, pp. 35-41. 35. Mo sdo , F.; Meie , E.; Kö z, B.; I en, K. I.; Dobbe in, M.; Allgöwe , B. LIDAR-based geome ic econs uc ion o bo eal ype o es s ands a single ee le el o o es and wildland i e managemen . Remo e Sens. En i on. 2004, 3, 353-362. Remo e Sens. 2012, 4 973 36. Wack, R.; Scha d , M.; Loh , U.; Ba ucho, L.; Oli ei a, T. Fo es In en o y o Eucalyp us Plan a ions Based on Ai bo ne Lase Scanne Da a. In P oceedings o ISPRS Wo kshop 3-D Recons uc ion om Ai bo ne Lase scanne and InSAR Da a, D esden, Ge many, 8–10 Oc obe 2003; In In e na ional A chi es o he Pho og amme y, Remo e Sensing and Spa ial In o ma ion Sciences; ISPRS: Vienna, Aus ia, 2003; Volume 34, Pa 3/W13, pp. 40-46. 37. Pi känen, J.; Mal amo, M.; Hyyppä, J.; Yu, X. Adap i e Me hods o Indi idual T ee De ec ion on Ai bo ne Lase Based Canopy Heigh Model. In P oceedings o ISPRS Wo kshop Lase -Scanne s o Fo es and Landscape Assessmen , F eibu g, Ge many, 3–6 Oc obe 2004; In In e na ional A chi es o Pho og amme y, Remo e Sensing and Spa ial In o ma ion Sciences; ISPRS: Vienna, Aus ia, 2004; Volume 36, Pa 8/W2, pp. 187-191. 38. Yu, X.; Hyyppä, J.; Vas a an a, M.; Holopainen, M.; Vii ala, R. P edic ing indi idual ee a ibu es om ai bo ne lase poin clouds based on andom o es s echnique. ISPRS J. Pho og amm. 2011, 66, 28-37. 39. Heinzel, J.N.; Weinacke , H.; Koch, B. P io -knowledge-based single- ee ex ac ion. In . J. Remo e Sens. 2011, 32, 4999-5020. 40. Vauhkonen, J.; Ene, L.; Gup a, S.; Heinzel, J.; Holmg en, J.; Pi känen, J.; Solbe g, S.; Wang; Y.; Weinacke , H.; Hauglin, K.M.; e al. Compa a i e es ing o single- ee de ec ion algo i hms unde di e en ypes o o es . Fo es y 2011, doi: 10.1093/ o es y/cp 051. 41. Peuhku inen, J.; Mal amo, M.; Malinen, J.; Pi känen, J.; Packalén, P. P eha es measu emen o ma ked s ands using ai bo ne lase scanning. Fo es Sci. 2007, 53, 653-661. 42. Falkowski, M.J.; Smi h, A.M.S.; Gessle , P.E.; Hudak, A.T.; Vie ling, L.A.; E ans, J.S. The in luence o coni e o es canopy co e on he accu acy o wo indi idual ee measu emen algo i hms using lida da a. Can. J. Remo e Sens. 2008, 34, 338-350. 43. Vas a an a, M.; Kanka e, V.; Holopainen, M.; Yu, X.; Hyyppä, J.; Hyyppä, H. Combina ion o indi idual ee de ec ion and a ea-based app oach in impu a ion o o es a iables using ai bo ne lase da a. ISPRS J. Pho og amm. 2012, 67, 73-79. 44. Axelsson, P. G ound Es ima ion o Lase Da a Using Adap i e TIN Models. In P oceedings o OEEPE Wo kshop on Ai bo ne Lase scanning and In e e ome ic SAR o De ailed Digi al Ele a ion Models, S ockholm, Sweden, 1–3 Ma ch 2001; Publica ion No. 40; pp. 185-208. 45. Bilke , M.; Kaa inen, H. The Quali y o Real-Time Kinema ic (RTK) GPS Posi ioning; Repo s o he Finnish Geode ic Ins i u e; FGI: Masala, Finland, 2001; pp. 1-25. 46. Wol (né S aub), B.-M.; Heipke, C. Au oma ic ex ac ion and delinea ion o single ees om emo e sensing da a. Machine Vis. Appl. 2007, 18, 317-330. 47. Solbe g, S.; Næsse , E.; Bollandsås, O.M. Single- ee segmen a ion using ai bo ne lase scanne da a in a s uc u ally he e ogeneous sp uce o es . Pho og amm. Eng. Rem. Sensing 2006, 72, 1369-1378. 48. Se a, J. Image Analysis and Ma hema ical Mo phology; Academic P ess: London, UK, 1982; Volume 1. 49. Se a, J. Image Analysis and Ma hema ical Mo phology: Theo e ical Ad ances; Academic P ess: London, UK, 1988; Volume 2. Remo e Sens. 2012, 4 974 50. Mal amo, M.; Peuhku inen, J.; Malinen, J.; Vauhkonen, J.; Packalén, P.; Tokola, T. P edic ing ee a ibu es and quali y cha ac e is ics o Sco s pine using ai bo ne lase scanning da a. Sil a Fennica 2009, 43, 507-521. 51. Vauhkonen, J.; Ko pela, I.; Mal amo, M.; Tokola, T. Impu a ion o single- ee a ibu es using ai bo ne lase scanning-based heigh , in ensi y, and alpha shape me ics. Remo e Sens. En i on. 2010, 114, 1263-1276. 52. Hyyppä, J.; Yu, X.; Hyyppä, H.; Vas a an a, M.; Holopainen, M.; Kukko, A.; Kaa inen, H.; Jaakkola, A.; Vaaja, M.; Koskinen, J.; Alho, A. Imp o ed o es in en o y using pene a ed Hi s and da a usion a ea u e le el. Remo e Sens. 2012, submi ed. 53. Hyyppä, J.; Mielonen, T.; Hyyppä, H.; Mal amo, M.; Honka aa a, E.; Yu, X.; Kaa inen, H. Using Indi idual T ee C own App oach o Fo es Volume Ex ac ion wi h Ae ial Images and Lase Poin Clouds. In P oceedings o The ISPRS Wo kshop Lase Scanning 2005, Enschede, The Ne he lands, 12–14 Sep embe 2005; In In e na ional A chi es o Pho og amme y, Remo e Sensing and Spa ial In o ma ion Sciences, ISPRS: Vienna, Aus ia, 2005; Volume 36, Pa 3/W19, pp. 144-149. 54. Hyyppä, J.; Yu, X.; Hyyppä, H.; Mal amo, M. Me hods o Ai bo ne Lase Scanning o Fo es In o ma ion Ex ac ion. P esen ed a Wo kshop on 3D Remo e Sensing in Fo es y, Vienna, Aus ia, 14–15 Feb ua y 2006; CD-ROM, p. 16 55. Lindbe g, E.; Holmg en, J.; Olo sson, K.; Walle man, J.; Olsson, H. Es ima ion o ee lis s om ai bo ne lase scanning by combining single- ee and a ea-based me hods. In . J. Remo e Sens. 2010, 31, 1175-1192. 56. B eidenbach, J.; Næsse , E.; Lien, V.; Gobakken, T.; Solbe g, S. P edic ion o species speci ic o es in en o y a ibu es using a nonpa ame ic semi-indi idual ee c own app oach based on used ai bo ne lase scanning and mul ispec al da a. Remo e Sens. En i on. 2010, 114, 911-924. 57. Vas a an a, M.; Holopainen, M.; Yu, X.; Haapanen, R.; Melkas, T.; Hyyppä, J.; Hyyppä. H. Indi idual ee de ec ion and a ea-based app oach in e ie al o o es in en o y cha ac e is ics om low-pulse ai bo ne lase scanning da a. Pho og amm. J. Fin. 2011, 22, 1-13. © 2012 by he au ho s; licensee MDPI, Basel, Swi ze land. This a icle is an open access a icle dis ibu ed unde he e ms and condi ions o he C ea i e Commons A ibu ion license (h p://c ea i ecommons.o g/licenses/by/3.0/).